Mapping Floods and Inundations in Urban Areas Using Machine Learning: An Approach Based on Press Reports with Natural and Anthropogenic Factors

Le résumé fourni par la source

Urban flooding occurs frequently and results in significant environmental and socioeconomic impacts. Addressing this issue begins with mapping areas susceptible to floods and inundations. This study focuses on the urban area of Campo Grande, Mato Grosso do Sul, Brazil, which recorded 1,100 flood-related points over a 9-year period. These occurrences were identified through reports published by two local digital news outlets, selected using keyword filters based on the study’s theme, temporal scope, and geographic location. To analyze flood susceptibility, we adopted a machine learning approach using the gradient boosting classifier (GBC). This method enabled the development of a predictive model based on the spatial distribution of flood events reported by the media, in combination with a set of input variables characterizing the study area. The objective was to estimate the number of flood and inundation events per year within the urban perimeter. The GBC model simulated an average of 173 occurrences per year in the area studied. These results highlight the potential of this technique as a tool for urban and environmental management. By identifying underreported areas most prone to flooding, the model can support the formulation of public policies, as well as the planning and implementation of mitigation and adaptation measures. This study demonstrates the feasibility and relevance of integrating alternative data sources—such as media reports—with artificial intelligence to improve flood risk mapping. The proposed methodology provides a replicable framework that can be applied in other urban contexts, particularly those with limited access to official monitoring data, reinforcing the value of interdisciplinary approaches in addressing urban environmental challenges.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.